Someone goes down to check the sump pump. Water is rising outside but the pump is holding, so they stay — to watch it, to get the boxes off the floor, to shove another bag against the foundation wall. Then the water comes faster than the pump can move it. The stairs are under before they turn around.
Twenty-two percent of the people who died in the July 2021 floods across Belgium and Germany died something like that: in basements, checking pumps, moving belongings, trying to hold down the damage.
We know this because a reconstruction published this year by Thieken and colleagues in Earth's Future asked a question flood fatality research usually leaves alone. The count was 224. The reconstruction went after the circumstances behind each one. Which floor of the building. What the person was doing when the water reached them. Whether they had been warned. Whether they had time to leave. Whether the place they drowned appeared on any official flood hazard map.
Of the 131 people who died indoors, two-thirds were caught by surprise: in basement apartments, or more often on ground and upper floors that the study's authors conclude should have been evacuated earlier. The water got into places, and to heights, that people had not expected.
And 58 percent of all 224 deaths happened outside officially mapped flood zones. On paper, those locations were safe.
Each finding marks a distance between how flood risk gets managed and how people actually drown. The question for an American reader is whether these are European patterns, or whether the United States simply lacks the categories to find them.
Where the maps end
The 58 percent travels fastest and needs the most care.
Flood maps in the United States do regulatory work. FEMA's Special Flood Hazard Area, commonly called the "100-year floodplain," is the line a lender checks before requiring flood insurance, the line a building code uses to set a minimum first-floor elevation, the line that determines which communities qualify for certain kinds of disaster assistance. As I wrote in an earlier piece about the 100-year label, the zone answers a regulatory question. It was never built to describe the whole geography of where a person can drown.
No national estimate exists for how many U.S. flood deaths fall outside that line. The large national datasets, including Han and Sharif's 61-year database and Ashley and Ashley's earlier analysis, sort deaths by activity and setting, not by whether the location was mapped. The NWS annual fatality tables don't do it either.
What exists is event reconstructions, and they disagree.
Jonkman and colleagues reconstructed 70 deaths from Hurricane Harvey in 2017 and found 80 percent outside the 100-year floodplain, 53 percent outside even the 500-year zone. Harvey was rain landing on a metropolitan drainage system that had been losing capacity to pavement for decades, a mechanism FEMA's riverine and coastal maps were not drawn to represent.
A West Virginia analysis of the state's June 2016 flood found the reverse: 91 percent of 23 deaths inside the FEMA 100-year zone. That water ran down narrow Appalachian valleys, where the mapped floodplain and the path the flood actually took are nearly the same shape.
The relationship between mapped risk and actual death depends on the kind of flood, the terrain it moves through, and what has been built on that terrain, all of which change from event to event and region to region.
The remnants of Ida in New York City, in September 2021, give a third configuration. Eleven people drowned in basement apartments across five homes. In congressional testimony, Representative Carolyn Maloney said all five homes stood in areas FEMA had designated as having minimal flood hazard. FEMA Administrator Deanne Criswell gave the reason: the maps accounted for coastal and river flooding, not for storm sewers backing up into below-grade rooms.
The European 58 percent doesn't convert into an American number, and nobody should try to make it. What carries across is the shape of the failure. Each map represents the flood mechanisms it was built to represent, and people keep dying from the ones it wasn't.
The basement
Thieken's 22 percent is the finding that presses hardest on preparation behavior. These people were not caught unaware in their sleep. They went down there on purpose. They had pumps, sandbags, the equipment the guidance tells you to have, installed where it needs to be. They were operating it when the water arrived faster than they had figured.
The American evidence for that specific pathway is thin, and the thinness is itself informative.
Han and Sharif's national database assigns exactly two deaths across 61 years to an equipment category that includes pump failure. The NWS does not record floor level or activity at time of death. If Americans are drowning while running sump pumps and stacking bags against basement walls, the national data system has no field for it.
Ida remains the strongest U.S. basement case, and it describes a different situation. Eight of the eleven drownings happened between midnight and 3 a.m. Investigators identified rapid water arrival, blocked exits, and structural barriers to rescue. Those were people trapped in below-grade apartments, many of them unregulated units: entrapment at home rather than work performed to protect property.
Scattered American cases do match the European pattern. NOAA Storm Data for April 1996 records a 74-year-old man in southwestern Indiana who had a heart attack while working on his sump pump, collapsed, and drowned on the basement floor. The West Virginia analysis put nine of 23 deaths under "risky behavior," a heading that covered retrieving memorabilia from basements but also driving into floodwater and playing near creeks. The report doesn't separate them.
The Texas legislative investigation into the July 2025 Hill Country flood, which killed at least 135 people and was the deadliest U.S. flash flood since 1976, documented one camper who died after going back to a cabin for something. The Texas reports describe no pattern of basement property-protection deaths. The mechanism there was different: water rising fast in the dark, evacuation attempted too late or by vehicle.
So the European finding has individual American analogues and no American frequency. Whether the pattern is genuinely rarer here or simply invisible to the counting is open. Building stock differs between the U.S. and Europe, as do basement prevalence and the dominant flood types. Nobody is asking the question systematically.
Caught indoors
Thieken documented a second indoor pathway, distinct from the basement: people who died on ground floors and upper floors, in rooms the study's authors conclude should have been evacuated earlier. These were living spaces, and the water got higher than the people inside them expected.
The U.S. system sees this even less well than it sees the basement. Han and Sharif recorded 216 deaths in permanent homes across 61 years, about 5.7 percent of deaths where circumstances were known. Whether those people had been warned, and whether there had been a window in which leaving was possible, is not in the record. You get the location and nothing about what preceded it.
Event reconstructions offer fragments. The West Virginia analysis found 74 percent of the state's 2016 flood deaths associated with private residential structures, and 87 percent of victims experiencing rapid water rise combined with failure to evacuate in time. But "failure to evacuate in time" holds two different situations: people who had no opportunity, and people who had one and didn't take it. The Camp Mystic investigation concluded there had been sufficient time to walk every camper out, and that the camp never activated its public-address system. Campers sheltered where they were, waiting for instructions that didn't come.
Which situation you are looking at changes what would have helped. No opportunity is a problem of warning lead time or physical access. Opportunity declined is a problem of risk perception, or attachment to the house, or distrust of the warning, or the memory of three previous floods that came up two inches and stopped. The national data doesn't sort those apart.
Why the U.S. data looks different
Both indoor questions are hard to answer here partly because of what fills the American statistics instead.
In Han and Sharif's database, vehicles account for 57.8 percent of flood deaths with known circumstances between 1959 and 2019. Add people who walked into floodwater and the two categories together come to 86 percent.
Recent NWS tables show a smaller vehicle share, 50 percent for 2015–2019 and 34 percent for 2020–2024, but the annual figures swing hard depending on which disasters land in a given year, and the peer-reviewed literature hasn't established a durable behavioral shift.
A number that size organizes everything downstream of it. It sets where research attention goes, what the safety campaigns say ("Turn Around, Don't Drown"), and which categories the data system bothers to resolve. Indoor deaths, basement deaths, deaths during property-protection work, deaths among people who stayed put: these are residuals in a system built around its dominant mechanism.
Thieken's study comes out of a place where the dominant mechanism is something else. In July 2021 the water came into the buildings and stayed. People died indoors, so the reconstruction was designed to ask indoor questions: floor, activity, warning, decision. American flood fatality research has mostly not been built that way, because American flood death has mostly happened in vehicles and in moving water outdoors.
People do die indoors here. The system that counts them was not built to record what would make those deaths legible.
The gap in the guidance
Set the numbers aside and read what federal agencies tell people to do before a flood.
The NWS advises residents to check that their sump pump works and to install a battery-operated backup. It recommends sandbags or other barriers "if you have sufficient time." FEMA's FloodSmart guidance tells homeowners to elevate utilities, clear drains, move furniture and valuables to higher floors, and keep a working sump pump with battery backup. Ready.gov says much the same.
Once a flood warning is issued, the messaging changes: higher ground, follow evacuation orders, don't walk or drive through floodwater.
The gap is in the handoff. The guidance tells you to install the pump and sandbag the foundation before the water comes. It tells you to get out when the water arrives. It does not tell you when to stop pumping, or when to leave the bags half-placed and walk up the stairs. In practice, the trigger for abandoning the work is an evacuation order or dangerous water at the door. But the pump and the sandbags exist to hold the water off, which puts the person running them in the basement exactly while conditions turn.
The standard NFIP policy reimburses up to $1,000 for loss-avoidance measures — sandbags, pumps, plastic sheeting, and labor valued at the federal minimum wage — when flooding is imminent enough that a reasonable person would anticipate it. The claims checklist leads with safety and tells policyholders not to reenter until authorities declare it safe. So the program defines a category of labor performed while flood conditions are developing, puts a price on it, and instructs the person performing it to prioritize staying alive. Where the first instruction ends and the second begins is left to whoever is standing on the basement stairs.
Many of the people Thieken's team reconstructed were doing that labor when they died. They were acting on the warnings, with the equipment the guidance told them to own, in the place that equipment is installed.
What the system cannot ask
The United States counts flood deaths. What it does not do at the national level is record where those deaths fall relative to mapped flood zones, which floor the victim was on, what the victim was doing when the water reached them, or whether they had received a warning and stayed anyway. The fields that would make the Thieken comparison possible are not standard in the NWS fatality database.
Circumstance data was missing for 41 percent of deaths in the 61-year national study. Age was missing for 63 percent. Sex for 58 percent.
Single events establish that each of these things happens here. Harvey put deaths outside the mapped zone; Ida put them in basements. A Storm Data entry from 1996 has a man dead on his own basement floor beside the pump he was working on. Camp Mystic had a group with time to leave and nobody to tell them to go. None of it gives a rate.
Thieken's reconstruction laid out a sequence: the map said the place was safe, the water reached floors the residents did not expect it to reach, the person went down to protect property or stayed upstairs while conditions worsened, and they did not come back up or get out. How common that sequence is in the United States is not something the existing data can answer. It is not something the existing data was built to ask.
If the preparation guidance and the fatal exposure overlap, and the national count cannot see the overlap, then the people inside it are there because they took the warnings seriously.
- Flood maps and property risk: A 2022 modeling study in Nature Climate Change estimated that only 41 percent of U.S. annual flood loss risk falls within FEMA's Special Flood Hazard Area, measuring property damage rather than fatalities but reinforcing the question of what the mapped boundary excludes.
- Adaptation labor as exposure: A panel discussion among disaster researchers and emergency practitioners explored how protective equipment — pumps, generators, flood barriers, air purifiers — creates monitoring and maintenance obligations that can place the operator in danger precisely when conditions worsen, a pattern the Thieken basement deaths illustrate at its most extreme.
- The July 2025 Texas toll: A NOAA-affiliated hydrometeorological reconstruction of the July 2025 South Texas floods reported 137 fatalities, but no published analysis has yet overlaid each death location against the FEMA floodplain or reconstructed individual fatal sequences the way the European study did.
- When the all-clear changes behavior: Research on river heatwaves lasting days beyond atmospheric cooling raises a parallel question about flood recovery — whether declaring a flood event over shifts behavior and withdraws attention while residual risks from contaminated water, structural damage, or saturated soils persist.